Toward Secure Task Offloading in Vehicular Edge Computing Networks

Vehicular Edge Computing (VEC) enables vehicular services but still struggles to achieve secure and efficient task offloading under highly dynamic and resource-constrained conditions. To tackle this problem, we develop a secure task offloading framework where task vehicles (TVs) determine offloading decisions while eavesdropping vehicles (EVs) attempt to intercept sensitive data. The joint optimization of offloading ratio, spectrum allocation, and transmission power is modeled as a cooperative multi-agent Markov decision process (MDP) that captures mobility-driven channel variations and resource competition. We then design a Deep Deterministic Policy Gradient (DDPG)-based Secure Task Offloading (DSTO) scheme to solve this continuous-control problem. DSTO encodes vehicular states into compact representations and uses an actor–critic (AC) architecture to learn adaptive strategies. The actor outputs continuous offloading and power-control actions, while the critic evaluates long-term utility under integrated security, delay, and energy constraints. Security is explicitly incorporated through the security rate, reflecting the legitimate link's advantage over the eavesdropping link. Leveraging DDPG's capability in high-dimensional continuous spaces, DSTO robustly adapts to dynamic vehicular conditions and balances security with system performance. Extensive experiments show that DSTO significantly reduces task delay and energy consumption, achieving over 12.7% improvement in comprehensive utility compared with baseline approaches.